Population

The study cohort comprised all Danish employees aged 20–69 years at 1 January 2020 (N=2 451 542). It was identified from records in the Work Classification Module at Statistics Denmark, as a subset of the Danish Occupational Cohort with eXposure data (DOC*X) (33). Other persons from the same households as cohort members (N=1 503 892, 74% ≤20 years of age; 1.4% ≥70 years of age) were identified through the residential address in order to account for viral transmission in the family. Permissions to remotely retrieve, compile and analyze pseudonymized data at a secured and logged platform at Statistics Denmark were obtained from the Danish Data Protection Agency (P-2020-897), Statistics Denmark (P-708121) and the National Board of Health Data (FSEID-00005368). This process started in August 2020, and access to the data-files was provided in December 2021.

Occupation and industry

Occupation and industry codes are primarily provided by employing companies but other sources of information such as tax records, trade union membership, and educational records are also used by Statistic Denmark. This allowed assignment of job titles according to the Danish version of the International Standard Classification of Occupations (DISCO-08) (34), for 86% at the 4-digit level, and assignment of industry codes according to the Danish version of the Statistical Classification of Economic Activities in the European Communities (DB07) (35) for 100% at the 2-digit level. When we retrieved data for this study in December 2021, the most updated information on occupational and economic activity codes was from December 2019. Therefore, we did not have data on changes of occupation in 2020 and 2021. With few exceptions, the Danish classifications are identical to the corresponding international classifications (ISCO-08 and NACE Rev. 2).

With one exception, it was not possible to infer missing DISCO-08 codes from known DB07 codes because individual DB07 codes are associated with a wide miscellany of DISCO-08 codes. However, as 75% of employees within ‘hairdressing and other beauty treatment’ (DB07 code 96.02.10) were hairdressers (DISCO-08 code 5141), all individuals with a missing DISCO-08 code within this industry (N=1135) were assigned the 5141 DISCO-08 code.

Outcome ascertainment

Data on admission to hospital for COVID-19-related disease from 1 January 2020 through 14 December 2021 were obtained from a daily updated database with national coverage kept at Statens Serum Institut, Copenhagen. A hospital admission was defined as COVID-19 related if a SARS-CoV-2 PCR swab test was positive up to 14 days prior to admission, and if the hospital stay was >12 hours (an algorithm used by Statens Serum Institut). An examination of ICD-10 diagnoses available from the National Patient Registry indicated that about 2.5% of cases using this definition were probably attributable to other causes such as psychiatric, traumatic, or obstetric disorders.

Reference group

The reference group was defined a priori as occupations classified to the lowest level of potential occupational exposure to SARS-CoV-2 by an expert-rated COVID-19 JEM (31). Eight domains addressing four factors on transmission risk (type, number, proximity, and location of social contacts at work), two mitigation measures (social distancing and face covering) and two job insecurity factors (income insecurity and proportion of migrants) were each rated on a scale from 0 (low risk of exposure) to 3 (high risk of exposure). The reference group was defined by the 50 job codes (4-digit DISCO-08 level) with low risk on each of the eight dimensions (sumscore_1-8 = 0, N=369 341, 15% of the study cohort). The three largest job groups in the reference population were general office clerks, commercial sales representatives, and accounting associate professionals (41% of the reference group, DISCO-08 codes 4110, 3322 and 3313, respectively).

An overview of the subsets of the study population is displayed in table 1.

Covariates

Using their unique personal identification numbers, we linked each cohort member with records in public registers hosted by Statistics Denmark to obtain information on a range of demographic, social and health variables at the end of 2019: sex, age, duration of education in years, country of origin, social position, household income, hospital admission for ≥1 of 11 chronic diseases during 2010–2019, geographical residential area and living in a municipality with a large city.

From data on household members, we defined variables indicating the size of the household sharing the same residence. The number of family members (excluding the index-person), who had ≥1 PCR swab test during the second or third week before any particular week during follow-up, and the results of those tests, were retrieved from the national testing database, which by mid December 2021 included results of about 40 million PCR tests. Tests were available free of charge throughout the country from early on in the first epidemic wave. The 2–3 week interval was specified to account for incubation period and disease development before hospitalization

Individual proxies of lifestyle factors in terms of sex-, age- and period-specific probability of current smoking and estimates of body mass index (kg/m²) were assigned by lifestyle JEM based on questionnaire information from several large random samples of the Danish population representative for 2010 (36).

From the DREAM register with longitudinal data on public benefit transfer payments to all Danish citizens (37), we obtained data on retirement each week during 2020 (data not available for 2021). Finally, we obtained information on the date of second vaccination against COVID-19 (if any) and vital status during 2020 and 2021.

Statistical analysis

We examined incident hospital admission for COVID-19 related disease as a function of occupation and covariates. Crude and adjusted measures of association, in terms of incidence rate ratios (IRR) with 95% confidence intervals (CI), were computed by Poisson regression applied to counts of COVID-19 hospital admissions and weeks at-risk for all covariate cross tabulations. Follow-up started in week 8 of 2020 (the week before first hospital admissions for COVID-19 in Denmark) and continued until (and including) week 50 of 2021 (the last week with available data on COVID-19 admissions). Follow-up was censored at the first of: COVID-19 related hospital admission (N=4416); death (N=2615); emigration (N=18 112); retirement (N=33 427); or week 50 of 2021.

A power calculation based upon the outcome occurrence in the reference group indicated that a relative risk of 1.8 would be detectable in an occupation with 80% power at the 5% significance level if it included >2000 employees. Therefore, we report risk estimates for the non-referent occupations of this size or larger, while keeping all jobs regardless of size in the reference group. Occupations with ≤2000 employees and employees with missing DISCO-08 codes were analyzed as separate categories.

Risk estimates by industrial sector. To account for possible differences between the same occupation in different industrial sectors – for instance cleaners in hospitals versus cleaners in office buildings and nursing aides in institutions versus nursing aides in private homes – we present occupational risk stratified by industrial sectors with more likely exposure to SARS-CoV-2. These industrial sectors were selected by an average COVID-19 JEM sum score >12 (the median) for industries classified at the 2-digit DB07 level. DB07 groups were pooled when appropriate, considering type of industrial sector (for instance land-, sea and air transportation, DB07 codes 49, 50 and 51). In few instances we deviated from the 4-digit DISCO-08 classification by splitting the 4-digit groups into 6-digit groups to distinguish more specific occupations (for instance day care assistants by age of the children taken care of) or to pool 4-digit groups into 3-digit groups to obtain sufficient sample size – for instance generalist and specialized medical practitioners or waiters/bartenders (explicitly detailed in table 2). Separate risk estimates are reported for occupations with a COVID-19 JEM sum score >12 but for completeness a summary risk estimate for all occupations with COVID-19 sum scores ≤12 is also provided.

Risk estimates across all industrial sectors. Using the average JEM sum score to select industrial sectors and, within these, occupations of particular interest, there is a risk of leaving out occupations that score high on some COVID-JEM dimensions (for instance number of daily contacts and work with the general public), while scoring low on other dimensions (for instance job insecurity). For that reason and to provide a full systematic quantification of occupational risk independent of industrial sector and COVID-19 JEM assessment, we estimated the risk for all non-referent occupations with >2000 employees at the 4-digit DISCO-08 level while keeping occupations with ≤2000 and employees with missing DISCO-08 data as separate categories.

Adjustments. In all analyses, we first adjusted the risk estimate by sex and age, which are strongly related to both occupation and outcome (being a male and higher age conferring a higher risk). Secondly, we further adjusted for a fixed set of baseline variables according to the disjunctive confounder variable selection criteria (38). Included covariates are listed in the footnote to table 2 which also displays grouping categories. Social position, household income, number of children in the household and living in a large city were also considered as possible confounding factors but were not included in final statistical models because of strong overlap with other covariates within the same domain. Retirement, death, and emigration during follow-up were accounted for by censoring, while COVID-19 vaccination was controlled for by a time-varying variable.

All analyses were carried out in SAS 9.4 (SAS Institute, Cary, NC, USA).

Occupational risk by industrial sector

We identified six major industrial sectors defined by the 2-digit DB07 level with an average COVID-19 JEM sum score >12. Increased risk of COVID-19 related hospital admission was observed in one or more 4-digit DISCO-08 occupations within three of these (healthcare, social care and transportation), but not in education, retail sales and a heterogeneous group of service jobs) (table 2).

Within healthcare, eight of thirteen occupations with ≥2000 employees had an increased risk, with health care assistants experiencing the highest risk, and physiotherapists, midwives, dentists and dental assistants as notable exceptions (table 2).

In social care, four out of fourteen occupations had increased risk, the highest IRR being in nurses. Nursing aides (in institutions and private homes, respectively) and teachers/day care assistants for children aged 4–7 also had increased risk, while primary school teachers and social work and counselling professionals had not.

Among five jobs in the transportation sector, increased risk was seen among bus and possibly taxi drivers but not lorry and locomotive engine drivers.

Occupational risk across all industrial sectors

Computation of IRR by 4-digit DISCO-08 occupations across all industrial sectors independently of COVID-19 JEM assessments revealed few additional occupations with elevated risk not captured by the industrial sector stratified analysis. These include construction managers, medical imaging and equipment operators, chefs, early childhood educators and food and related products machine operators (table 3 and supplementary material www.sjweh.fi/article/4063, table S1). Moreover, low risk was seen in seven occupations, mainly some high-skill and blue-collar occupations (supplementary table S1).

Supplementary results

As expected, several demographic and social characteristics were unevenly distributed across categories of the COVID-19 JEM sum score, in particular, sex, age, duration of education and country of birth. Differences were less pronounced regarding size of household, health, and lifestyle (supplementary table S2).

Strong and robust sex- and age-adjusted associations were seen between COVID-19 related hospital admission and most covariates other than body mass index and positive PCR swab test of one or more family members in the previous 2–3 weeks (supplementary table S3)

Methodological issues

Hospital admission is not a perfect proxy for incidence of COVID-19. Admission rates among infected individuals are higher among those who are older, male, of non-white ethnicity, with various comorbidities, and have an unhealthy lifestyle (39, 40). To reduce any resultant bias, we adjusted risk estimates for age, sex, country of origin, history of hospital admission for other diseases, body mass index and tobacco smoking. Immunization, once it became available, will have reduced individual vulnerability to more severe infection requiring hospital admission, and therefore was a further factor of adjustment. There is also potential for confounding by non-occupational exposure to COVID-19. To reduce this bias, we adjusted for years of education, household size, and recent documentation of infection in a household member. A further complication is that during the study period, incidence of COVID-19 varied by region within Denmark, which could lead to bias where jobs were not uniformly distributed across the country. Therefore, we included adjustment for region. Despite these many adjustments, we cannot exclude some residual bias – for instance, we had no data on commuting to work by public transport.

A large number of comparisons inevitably produces chance associations, and real associations cannot be distinguished from spurious associations on statistical grounds. The main approach taken to address the multi-comparison problem was to focus on occupations that according to independent expert ratings a priori were more likely to confer increased risk of viral infection. Nevertheless, it should be acknowledged that demonstration of increased (or reduced) risk in some occupations could be the result of random variation regardless of statistical significance. Cautious interpretation and critical assessment of plausibility are needed before specific results are communicated or actions taken based upon findings related to specific occupations. The detail provided by stratification of jobs by industrial sector may be helpful in risk assessment because we expect that the same jobs in different sectors will in some cases have different risks. On the other hand, real risks in some occupations may have been missed because of small numbers. It was, for example, unexpected that the risk in midwives was not elevated, although an explanation might be that both pregnant women and midwives have been unusually careful to avoid infection.

Positive PCR swab tests among family members were only weak predictors of COVID-19 related hospital admission 2–3 weeks later despite an extremely high testing activity, particularly during the second year of the epidemic. This finding cannot be taken as strong evidence that viral transmission in the family is of minor importance. Since positive PCR tests are strongly clustered in families in time – many performed within hours the very same day – the sequence of positive testing in a family may not be a reliable indicator of the sequence of infection. Moreover, in spite of high test-activity, there may still be many undetected subclinical cases.

The follow-up period included the first two epidemic waves and the initial part of the third wave. The end of follow-up coincided with a time when the less virulent omicron virus variant outnumbered the more lethal alpha, beta, and delta variants. However, there is no reason to expect that this change will have affected occupations differentially, and therefore it should not have been a source of bias.

Context and implications of findings

When comparing results across studies undertaken in different places and over different time periods, there is a possibility that differences in risk estimates could reflect not only the effects of bias and chance, but also variation in the extent of societal lockdown and effectiveness of precautionary measures as implemented in different countries at different stages of the pandemic. Results reported here may not apply to other countries or other time periods. Nevertheless, our findings are broadly consistent with those from elsewhere. They corroborate several earlier reports of increased risk among healthcare workers (1, 5–9, 12–14, 16, 21, 23–27), adding information on risk in various specific occupations not previously addressed in this sector – for instance medical laboratory technicians, hospital porters and psychologists. Increased risk among bus drivers has also been reported previously (18, 19). Elevated risk among shop assistants was observed in a mortality study (17) but not in our study. The absence of increased risk among teachers is notable, and again accords with other investigations (4, 41). That risk was increased in day care assistants providing daytime care for 4–7-year-old children might reflect closer contacts than experienced by primary school teachers, or that the latter were working from home when the epidemic peaked in Denmark in spring 2020 and winter 2021–22.

To the extent that the observed risk estimates reflect viral transmission related to working in a particular occupation, our data indicate that in the population and time period studied, recommendations and precautions to mitigate risk were insufficient and need to be revisited by employers and the Labour Inspection Agencies. When working at home is infeasible during periods with high community prevalence of infection, there is a need to develop and implement job- and task-specific guidelines based on the well-established general recommendations. At the same time, the demonstration of increased risk of infection in certain occupations may suggest interventions to protect the public more widely. For instance, elevation of risk among hairdressers might imply an increased risk for customers and family members too. As yet, however, there is little direct evidence of onward household transmission from people working in high-risk occupations (28, 42).

In most occupations where no or only marginally elevated risks (IRR <1.1) were observed, it seems likely either that viral transmission was not occurring in the workplace or that precautions were sufficient to mitigate it – as for instance in the travel, hotel and restaurant businesses, which have been locked down for long periods during the pandemic.

Significantly low risk estimates may have occurred in some jobs because limited contact with other people at work meant that SARS-CoV-2 transmission was lower than during leisure-time activities. Alternatively, they could be a consequence of random sampling variation in a context of multiple testing, or the reference groups used may not adequately have represented the lowest occupational exposure levels. The COVID-19 JEM was based upon subjective expert ratings and has so far not been validated (31).

The excessive societal impact of the COVID-19 pandemic may have consequences for occupational health reaching beyond the pandemic itself. In general, transmission of infectious diseases in the workplace has not been given high priority either in research agendas or in occupational health prevention programs (43, 44). This situation my change as documentation of the impact of workplace transmission of COVID-19 becomes stronger and widely accepted.

Concluding remarks

Our study provides strong evidence that the workplace has been an important setting for viral transmission during the SARS-CoV-2 pandemic in Denmark, and that lock-down and other measures to reduce such transmission have not been fully effective. Ways should be sought to reduce risk further, especially in health and social care, and among bus and taxi drivers. In addition, there is mounting evidence to support compensation for COVID-19 as an occupational disease in a number of jobs.